When should an AI care assistant ask before treating a missed routine as a problem?

A new preprint uses a missed medication as a small example of why “bring the pill” is a thin response: the person may have forgotten, declined, been confused, or be dealing with a side effect. It is a framework paper, not clinical evidence or a care protocol. But the design question feels real: when a home AI sees a missed routine, what should it ask or show before it turns a signal into a nudge, a reminder, or a message to someone else? I want a visible “not now,” “wrong assumption,” or “ask me later” before any escalation. A robot or speaker in the room should not turn a private choice into a compliance event.